Product Design for AI-Generated Email Summaries Built on Google's Gemini

You are a product manager at Google. Design an email summarization feature powered by Gemini, Google's family of AI models. ```hint Pick the email moment first Decide which situation the summary is for, such as catching up on a long thread or triaging a full inbox, and whose inbox it is, before you think about the model. ``` ### Clarifying Questions - Should the summarizer live inside Gmail only, or also work in other surfaces such as the Gemini app and for non-Gmail accounts? - Is the main audience consumer Gmail users, Google Workspace business customers, or both? - Should summaries appear automatically when an email or thread is opened, or only when the user asks for one? - What does Google want most from the launch: time saved for users, adoption of Gemini, or a reason to pay for a subscription that includes Gemini? ### What a Strong Answer Covers - A goal for the feature, plus the users and email situations it targets, chosen with a reason. - The summary experience: what gets summarized, when and where it appears, and how users act on it or check it against the original. - A high-level system view covering how email reaches the model, latency, the cost of summarizing at email scale, and caching. - How the design handles wrong or missing facts, the privacy of email content, and malicious emails written to manipulate the model. - Success metrics with guardrails, and a staged launch plan. ### Follow-up Questions - A summary leaves out a deadline mentioned in the fourth reply of a thread, and the user misses it. How would you detect and reduce this kind of failure? - An attacker sends an email with hidden text telling the model to warn the reader to call a phone number. How does your design stop it? - Summarizing every incoming email turns out to be too expensive. What do you cut first, and why?